corrected the voice conversion note, melody behaviour is not what the card said

gguf conversions, f16 through q4_k plus the experimental depth quantized variants

sample reference clip

model card

initial commit